Memory

Memory and privacy

You decide what survives. Three modes, switch any time with /privacy:

Mode	What is kept	Good for
ghost	nothing, ever	sensitive one-offs (phoenix run default)
session	this conversation only, gone on restart	everyday chat (default)
recall	notes you can read and audit	a real assistant that knows you

Recall mode

Phoenix keeps plain text files you can open, edit or delete:

• ~/.openphoenix/memory.md: long-term notes
• ~/.openphoenix/memory/2026-07-28.md: daily notes
• ~/.openphoenix/persona/: who your assistant is (Migration)

"What did we say about the server?" searches notes and past chat transcripts. Secrets are scrubbed before anything touches disk.

/wipe clears the current conversation, /forget drops a single remembered note, and deleting the files deletes the memory. No hidden databases.

Long conversations look after themselves

Chat all day and you will never hit a wall. When history approaches the model's limit, phoenix summarizes the older half and keeps going, preserving facts, decisions, file paths and open tasks. Really long histories are summarized in parts, so compaction still works when the backlog is larger than the model window itself.

Want it now, or want to steer it?

/compact
/compact focus on the deploy steps and open bugs

Prefer a fixed trigger? Set compact_after in [agent] to squash once history passes that many messages. Ghost mode never compacts, because it never keeps anything to begin with.

Dreaming

Idle with dreaming enabled? Phoenix journals its day to dreams.md. Harmless, charming, optional.